Decoding the Critique Mechanism in Large Reasoning Models

TL;DR

This paper investigates how large reasoning models detect and correct their own errors during complex reasoning tasks, revealing a hidden critique mechanism that improves their self-verification and error recovery abilities.

Contribution

It introduces the concept of a critique vector representing the model's internal error detection mechanism and demonstrates how steering this vector enhances model self-correction without additional training.

Findings

Models can reach correct answers despite intermediate errors.

Steering latent representations improves error detection.

The critique vector is highly interpretable and effective.

Abstract

Large Reasoning Models (LRMs) exhibit backtracking and self-verification mechanisms that enable them to revise intermediate steps and reach correct solutions, yielding strong performance on complex logical benchmarks. We hypothesize that such behaviors are beneficial only when the model has sufficiently strong "critique" ability to detect its own mistakes. This work systematically investigates how current LRMs recover from errors by inserting arithmetic mistakes in their intermediate reasoning steps. Notably, we discover a peculiar yet important phenomenon: despite the error propagating through the chain-of-thought (CoT), resulting in an incorrect intermediate conclusion, the model still reaches the correct final answer. This recovery implies that the model must possess an internal mechanism to detect errors and trigger self-correction, which we refer to as the hidden critique ability. Building on feature space analysis, we identify a highly interpretable critique vector representing this behavior. Extensive experiments across multiple model scales and families demonstrate that steering latent representations with this vector improves the model's error detection capability and enhances the performance of test-time scaling at no extra training cost. Our findings provide a valuable understanding of LRMs' critique behavior, suggesting a promising direction to control and improve their self-verification mechanism. Our code is available at https://github.com/mail-research/lrm-critique-vectors.